More than 80% of texts created by popular chatbots successfully pass the uniqueness test with a result of more than 75%. This was announced by scientists at Moscow State University named after Lomonosov.
Researchers tested four different services, which were asked to compose answers to written assignments from English textbooks. The texts were evaluated according to five criteria, including length, lexical diversity and level of plagiarism.
Large language models were selected, which, according to benchmarks from previous studies, show the best results in argumentation and reasoning. 80% of the generated texts passed the uniqueness test with a result of more than 75%
Scientists believe that working with such systems opens up new opportunities. With their help, you can process large amounts of data and get more creative results. Experts at Moscow State University believe that with the spread of large language models, there is a need to develop methods for adapting neural networks to current educational tasks.
An important area for further research is the study of the wording of the most successful queries (prompts) and, as a result, the identification of ways to form effective strategies for interaction with AI technologies within the professional competence of students
Read more on the topic:
A model of a computing system with AI was created at Moscow State University
MSU will develop a system for monitoring AI hallucinations
The problem of labor shortages in transport is planned to be solved with the help of AI
